Improved peri-operative outcomes with epidural analgesia in patients undergoing a pancreatectomy: a nationwide analysis

Transcription

1 DOI: /hpb HPB ORIGINAL ARTICLE Improved peri-operative outcomes with epidural analgesia in patients undergoing a pancreatectomy: a nationwide analysis Dominic E. Sanford 1, William G. Hawkins 1,2 & Ryan C. Fields 1,2 1 Department of Surgery, Barnes-Jewish Hospital and Washington University School of Medicine, and 2 Alvin J. Siteman Cancer Center, Washington University School of Medicine, Saint Louis, MO, USA Abstract Background: In spite of limited evidence demonstrating a benefit, epidural analgesia (EA) is often used for patients undergoing a pancreatectomy. In the present study, the impact of epidural analgesia on post-operative outcomes after a pancreatectomy is examined. Methods: Utilizing the Nationwide Inpatient Sample, the effect of EA on peri-operative outcomes after a pancreatectomy was examined. Multivariable logistic and linear regression with propensity score matching were utilized for risk adjustment. Results: From , patients underwent a pancreatectomy. Of these, 1130 (9.1%) patients received epidural analgesia. Using univariate comparison, patients receiving EA had a significantly decreased length of stay (LOS), hospital charges and post-operative inpatient mortality. In multivariate analyses, EA was independently associated with a decreased post-operative LOS (adjusted mean difference = 1.19 days, P < 0.001), decreased hospital charges (adjusted mean difference = $16 814, P = 0.002) and decreased post-operative inpatient mortality [adjusted odds ratio (OR) = 0.42, P < 0.001]. Using 1:1 propensity score matching, patients who received an EA (n = 1070) had significantly decreased post-operative LOS (11.0 versus 12.1 days, P = 0.011), lower hospital charges ($ versus $ , P = 0.001) and decreased post-operative inpatient mortality (1.5% versus 3.6%, P = 0.002) compared with matched controls without EA (n = 1070). Conclusion: Analysis of a large hospital database reveals that EA is associated with improved perioperative outcomes after a pancreatectomy. Additional studies are required to understand fully if this relationship is causal. Received 27 October 2014; accepted 16 December 2014 Correspondence Ryan C. Fields, Washington University in Saint Louis, 4990 Children s Place, Suite 1160, Box 8109, Saint Louis, MO 63110, USA. Tel: Fax: fieldsr@wudosis. wustl.edu Background Epidural Analgesia (EA) improves peri-operative outcomes in patients undergoing select operations. For example, EA has been associated with improved pain control as well as decreased post-operative respiratory failure, decreased postoperative pneumonia and quicker return of post-operative bowel function compared with standard methods of pain control. 1 6 A pancreatectomy is a morbid procedure with a high rate of complications, which influence post-operative length of stay (LOS), hospital cost and post-operative mortality. However, the role of EA in patients undergoing a pancreatectomy is poorly defined. Thus, further investigation regarding the role of EA in the care of patients undergoing a pancreatectomy is warranted. In spite of minimal evidence demonstrating the benefit after a pancreatectomy, EA is frequently used. Many pancreatic surgeons extrapolate data from studies using EA in other abdominal operations, and routinely use EA in their patients. With the growing use of enhanced recovery after surgery (ERAS) programmes in pancreatectomy patients, many of which incorporate EA, it is imperative to understand what role, if any, EA should play in post-operative care. 7,8 Thus, large multi-institutional studies would be ideal to help determine the impact of EA on post-operative outcomes in the pancreatectomy population. This study sought to examine the impact of EA on postoperative outcomes after a pancreatectomy on a national level. The Agency for Healthcare Research and Quality (AHRQ)

2 552 HPB Healthcare Cost and Utilization Project (HCUP) Nationwide Inpatient Sample (NIS) for 2008 to 2011 was utilized to determine whether EA had an effect on post-operative outcomes in patients undergoing a pancreatectomy. The hypothesis was that patients, who receive EA, have decreased post-operative LOS, decreased hospital charges and decreased post-operative inpatient mortality after risk-adjustment for patient, operative and hospital factors. Patients and methods Study design and patient population This was a retrospective cohort study using the AHRQ HCUP NIS for 2008 to 2011 to identify patients age 18 years or older undergoing a pancreaticoduodenectomy [International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9-CM) procedure codes: and 52.7], total pancreatectomy (ICD9-CM procedure code: 52.6) and distal pancreatectomy (ICD9-CM procedure codes: 52.52). Twenty-seven patient admissions were deleted because these were duplicate records. The NIS is the largest publicly available all-payer inpatient health care database in the United States, containing data from more than 7 million hospital stays each year. Study variables Common post-operative complications using ICD9-CM diagnosis codes in pancreatectomy patients have been identified in a similar AHRQ HCUP database. 9 Patients who received EA were identified by ICD-9-CM procedure codes and 03.91, as has been done previously using the NIS. 10 The AHRQ comorbidity software, Version 3.7, was used to identify comorbidities present at admission by utilizing measures defined by Elixhauser et al. based on ICD9-CM diagnosis codes. 11 The other race category included Asians, Pacific Islanders, Native Americans and Hispanics. Hospital pancreatectomy volume quartiles were defined using the total number of pancreatectomies performed at individual hospitals between 2008 and The American Hospital Association (AHA) Annual Survey of the Hospitals file was used to determine hospital bed size and teaching status. AHA hospital size is based on the number of hospital beds, specific to the hospital s location and teaching status. Cells with fewer than 11 patients per variable were relabelled as <11 in compliance with the HCUP data use agreement. Statistical analysis Chi-square and Student s t tests were used for univariate comparisons. Multivariable logistic and linear regression were used to examine the association of EA with LOS, hospital charges and inpatient mortality as appropriate. Propensity scores were estimated using a non-parsimonius multivariable logistic regression model including age, gender, race, comorbidities (congestive heart failure, chronic lung disease, diabetes, chronic renal failure, obesity, weight loss, alcohol abuse and drug abuse), insurance status, pancreatectomy type, cancer status, hospital pancreatectomy volume, hospital teaching status, and AHA hospital size with epidural analgesia as the dependent variable. 12 Patients who received EA were then matched 1:1 to patients who did not receive EA using a greedy matching algorithm with a caliper width of 0.2 standard deviations of the logit of the propensity score. 13 Covariate balance between matched pairs was assessed using the standardized difference, with values less than 10% indicating minimal imbalance. 14 All P-values were two-sided and values <0.05 were considered statistically significant in all analyses. All statistics were performed using SAS version 9.3 (SAS Inc. Cary, NC, USA). Results From 2009 through to 2011, patients underwent a pancreatectomy in the NIS. The mean age was 61.9 years (standard deviation: 13.8 years) and 51.0% were women. Patient, operative and hospital statistics are summarized in Table 1. Unadjusted outcomes by EA status after a pancreatectomy are shown in Table 2. After adjusting for age, gender, race, comorbidities, insurance status, pancreatectomy type, cancer status, hospital pancreatectomy volume, hospital teaching status and AHA hospital size (Table 3), EA was independently associated with decreased post-operative LOS, decreased hospital charges and decreased inpatient mortality. A 1:1 propensity score matching was performed using variables shown in Table 4. In all, 1070 matched pairs were obtained for the comparison, a match rate of 94.7% for all patients with epidurals (Table 4). The groups were well balanced with standardized differences of less than 10% for all variables. Patients, who received EA, had significantly improved outcomes compared with propensity score-matched controls without EA (Table 5). Discussion Previous studies examining the role of EA in pancreatectomy patients have failed to show a benefit. In a study of patients undergoing a pancreaticoduodenectomy, Pratt et al. 15 found that patients, who received EA, had lower pain scores, but also had increased rates of major complications. In a study including gastrectomy and pancreatectomy patients, Shah et al. 16 found that EA did not significantly improve pain control nor was it associated with significantly different rates of post-operative complications, such as pneumonia and ileus, or death. It is likely that the efficacy of epidural catheters is dependent on multiple factors. For example, the skill level and experience of the anaesthesiologist, the use of narcotics versus local anesthetics and how catheters are managed intra- and post-operatively all likely play a major role in whether EA is effective. 17 Thus, there may be tremendous institutional bias in a single institution studies regarding the efficacy of EA. A population-level

3 HPB 553 Table 1 Characteristics of pancreatectomy patients with and without an epidural analgesia Characteristic Mean (SD) or No. (Column %) P-value No Epidural (n = ) Epidural (n = 1130) Age, years Mean (SD) 62.0 (13.9) 61.6 (13.7) Gender Male 5530 (48.9%) 562 (49.7%) Female 5770 (51.0%) 564 (49.9%) Missing a <11 <11 Race White 7657 (67.7%) 706 (62.5%) <0.001 Black 972 (8.6%) 65 (5.8%) Other 2681 (23.7%) 359 (31.8%) Comorbidities CHF 368 (3.3%) 28 (2.5%) Chronic Lung Disease 1463 (12.9%) 154 (13.6%) Diabetes 3220 (28.5%) 323 (28.6%) Chronic renal failure 426 (3.8%) 40 (3.5%) Obesity 954 (8.4%) 80 (7.1%) Weight loss 1665 (14.7%) 192 (17.0%) Alcohol abuse 417 (3.7%) 37 (3.3%) Drug abuse 133 (1.2%) 19 (1.7%) Insurance Private 4647 (41.1%) 477 (42.2%) Medicaid 780 (6.9%) 61 (5.4%) Self-Pay 350 (3.1%) 36 (3.2%) Medicare 5158 (45.6%) 503 (44.5%) Missing 375 (3.3%) 53 (4.7%) Pancreatectomy type Whipple 7403 (65.5%) 838 (74.2%) <0.001 Total 589 (5.2%) 51 (4.5%) Distal 3318 (29.3%) 241 (21.3%) Diagnosis Malignant 7402 (65.5%) 775 (68.6%) Benign 3908 (34.6%) 355 (31.4%) Hospital pancreatectomy volume quartile 1st (<22 patients) 2894 (25.6%) 219 (19.4%) < nd (22 69 patients) 2837 (25.1%) 309 (27.4%) 3rd ( patients) 2869 (25.4%) 219 (19.4%) 4th (>139 patients) 2710 (24.0%) 383 (33.9%) Hospital teaching status Teaching 1819 (16.1%) 173 (15.3%) Non-teaching 9491 (83.9%) 957 (84.7%)

4 554 HPB Table 1 Continued Characteristic Mean (SD) or No. (Column %) P-value AHA hospital size b No Epidural (n = ) Epidural (n = 1130) Small 600 (5.3%) 56 (5.0%) Medium 1506 (13.3%) 149 (13.2%) Large 9081 (80.3%) 923 (81.7%) Missing a 123 (1.1%) <11 a Cells with fewer than 11 patients per variable were relabeled as <11 in compliance with the HCUP data use agreement. b AHA hospital size is based on the number of hospital beds, specific to the hospital s location and teaching status. Table 2 Unadjusted post-operative outcomes of pancreatectomy patients with and without an epidural analgesia Outcome Mean (SD) or No. (Column %) P value No epidural (n = ) Epidural (n = 1130) Postoperative length of stay, days 12.0 (10.5%) 10.9 (8.7) <0.001 Total hospital charges, $ ( ) ( ) <0.001 Inpatient mortality 391 (3.5%) 17 (1.5%) <0.001 Any complication 4902 (43.3%) 493 (43.6%) Myocardial infarction 105 (0.9%) 12 (1.1%) Hypotension/Shock 254 (2.3%) 14 (1.2%) Pneumonia/Respiratory failure 1138 (10.1%) 73 (6.5%) <0.001 Gastroparesis/nausea & vomiting 974 (8.6%) 110 (9.7%) Total parenteral nutrition 1421 (12.6%) 118 (10.4%) Acute renal failure 608 (5.4%) 38 (3.4%) Any infection 2681 (23.7%) 253 (22.4%) Urinary tract infection 784 (6.9%) 89 (7.9%) Surgical site infection 1216 (10.8%) 136 (12.0%) Blood transfusions 3042 (26.9%) 261 (23.1%) DVT/PE 280 (2.5%) 32 (2.8%) approach was used in an attempt to overcome this limitation, and avoid drawing conclusions based on the EA practices of a single institution. Previous randomized trials have unequivocally demonstrated a benefit of epidural analgesia over standard intravenous narcotics in patients undergoing abdominal operations about pain relief. 2 In addition, previous randomized studies and meta-analyses have demonstrated that EA is associated with decreased rates of certain complications, such as pulmonary morbidity and ileus, as well as decreased mortality. 5,6 Similarly, in the propensity score, matched analysis, patients who received EA had decreased rates of respiratory failure/pneumonia and death after a pancreatectomy. In addition, patients who received EA also had decreased rates of acute renal failure, an association that has also been reported with EA in other types of operations. 4,18 In multivariate analyses, pancreatectomies performed at hospitals in the lowest volume quartile were associated with increased LOS, increased hospital charges and increased inpatient mortality. Previous studies have found an association between lower pancreatectomy volume hospitals and worse peri-operative outcomes, 19,20 and this study confirms this finding. While EA use was more common at hospitals in the highest pancreatectomy volume quartile, the association of hospital pancreatectomy volume with improved post-operative outcomes was independent of epidural use. This suggests that the effect of increased volume on improved outcomes is not mediated solely by the use of EA. A major strength of this study was the ability to examine the impact of EA at the population level. Potential institutional confounders, such as hospital size, pancreatectomy volume and teaching status were accounted for in risk-adjustment, and a large number of patients were included, which provided the power to detect a difference in mortality. In addition to traditional multivariate regression methods, propensity score matching was used as a separate analysis to examine the impact of EA on post-operative outcomes. Because the decision for a patient to receive EA is largely dependent on surgeon preferences and institutional practices, it is not appropriate to directly compare patients who received EA to those

5 HPB 555 Table 3 Multivariate analyses of factors associated with post-operative outcomes after a pancreatectomy (n = ) Variable Length of stay, days Hospital charges, $ Inpatient mortality Mean difference (95% CI) P value Mean difference (95% CI) P-value OR (95% CI) P-value Age, per year 0.01 ( 0.00 to 0.03) ( 158 to 297) (1.02 to 1.05) <0.001 Sex Race Male reference Reference reference Female a 0.91 ( 1.29 to 0.55) < ( to 9 714) < (0.66 to 0.99) White reference Reference reference Black 0.45 ( 0.21 to 1.10) ( to 9 806) (0.61 to 1.38) Other 0.33 ( 0.15 to 0.81) (197 to ) (0.99 to 1.58) Comorbidities CHF a 3.78 (2.76 to 4.82) < ( to ) < (1.54 to 3.16) <0.001 Chronic lung disease 0.21 ( 0.75 to 0.33) ( to 4 019) (0.79 to 1.39) Diabetes 0.54 ( 0.43 to 1.52) ( to ) (0.63 to 1.63) Chronic renal failure a 1.35 (0.38 to 2.32) ( to ) < (1.41 to 2.95) <0.001 Obesity 0.00 ( 0.65 to 0.65) ( to 7 945) (0.81 to 1.69) Weight loss a 6.19 (5.67 to 6.72) < ( to ) < (1.51 to 2.37) <0.001 Alcohol abuse 0.37 ( 0.62 to 1.36) (668 to ) (0.82 to 2.34) Drug abuse 1.02 ( 0.67 to 2.71) ( to ) (0.48 to 3.20) Insurance Private reference reference reference Medicaid a 2.13 (1.37 to 2.89) < ( to ) < (1.08 to 2.62) Self-Pay 0.74 ( 0.32 to 1.80) ( to ) (0.82 to 2.83) Medicare 0.80 (0.33 to 1.28) (3 613 to ) (0.94 to 1.65) Pancreatectomy type Distal reference reference reference Whipple a 4.51 (4.27 to 5.15) < ( to ) < (1.94 to 3.72) <0.001 Total a 5.03 (4.14 to 5.92) < ( to ) < (1.65 to 4.83) <0.001 Diagnosis Benign reference reference reference Malignant 0.28 ( 0.70 to 0.15) (459 to ) (0.93 to 1.58) Hospital pancreatectomy volume quartile 1st (<22 patients) a 2.59 (1.95 to 3.23) < ( to ) < (1.61 to 3.19) < nd (22 69 patients) 1.53 (1.00 to 2.07) < ( to 8 036) (1.10 to 2.07) rd ( patients) 0.05 ( 0.48 to 0.58) ( to ) < (0.65 to 1.30) th (>139 patients) reference reference reference Hospital teaching status Teaching 0.74 (0.17 to 1.31) (7 608 to ) < (0.75 to 1.31) Non-teaching reference reference reference AHA hospital size b Small 0.27 ( 0.68 to 1.21) ( to ) < (0.51 to 1.29) Medium 0.01 ( 0.57 to 0.54) ( to 5 340) (0.64 to 1.16) Large reference reference reference Epidural a 1.19 ( 1.86 to 0.53) < ( to 8 912) (0.26 to 0.69) <0.001 a Denotes variables significantly associated with all three postoperative outcome measures. b AHA hospital size is based on the number of hospital beds, specific to the hospital s location and teaching status.

6 556 HPB Table 4 Propensity score matched comparison of patients with and without epidurals Characteristic Mean (SD) or No. (Column %) Standardized difference a Age, years Sex Race No epidural (n = 1070) Epidural (n = 1070) Mean (SD) 63.0 (13.2) 62.3 (13.2) 4.8% Male 531 (49.6%) 537 (50.2%) 1.1% Female 539 (50.4%) 533 (49.8%) White 669 (62.5%) 678 (63.4%) 3.2% Black 57 (5.3%) 62 (5.8%) Other 344 (32.2%) 330 (30.8%) Comorbidities CHF 27 (2.5%) 27 (2.5%) <1.0% Chronic lung disease 143 (13.4%) 147 (13.7%) 1.1% Diabetes 327 (30.6%) 306 (28.6%) 4.3% Chronic renal failure 29 (2.7%) 39 (3.6%) 5.3% Obesity 69 (6.5%) 74 (6.9%) 1.9% Weight loss 166 (15.5%) 171 (16.0%) 1.3% Alcohol abuse 26 (2.4%) 33 (3.1%) 4.0% Drug abuse 14 (1.3%) 17 (1.6%) 2.4% Insurance Private 474 (44.3%) 475 (44.4%) 4.4% Medicaid 55 (5.1%) 61 (5.7%) Self-Pay 29 (2.7%) 35 (3.3%) Medicare 512 (47.9%) 499 (46.6%) Pancreatectomy type Whipple 808 (75.5%) 804 (75.1%) 3.2% Total 38 (3.6%) 33 (3.1%) Distal 224 (20.9%) 233 (21.8%) Diagnosis Malignant 782 (73.1%) 751 (70.2%) 6.4% Benign 288 (26.9%) 319 (29.8%) Hospital pancreatectomy volume quartile 1st (<22 patients) 223 (20.8%) 211 (19.7%) 3.6% 2nd (22 69 patients) 292 (27.3%) 295 (27.6%) 3rd ( patients) 220 (20.6%) 215 (20.1%) 4th (>139 patients) 335 (31.3%) 349 (32.6%) Hospital teaching status Teaching 884 (82.6%) 902 (84.3%) 4.5% Non-teaching 186 (17.4%) 168 (15.7%) AHA hospital size b Small 49 (4.6%) 56 (5.2%) 3.8% Medium 136 (12.7%) 143 (13.4%) Large 885 (82.7%) 871 (81.4%) a Standardized Difference <10% indicates minimal covariate imbalance between propensity matched cohorts (Austin et al. Stat Med. 2009) b AHA hospital size is based on the number of hospital beds, specific to the hospital s location and teaching status.

7 HPB 557 Table 5 Post-operative outcomes after pancreatectomy of propensity score matched patients with and without epidural analgesia Outcome Mean (SD) or No. (Column %) P-value No epidural (n = 1070) Epidural (n = 1070) Postoperative length of stay, days 12.1 (10.8) 11.0 (10.4) Total hospital charges, $ ( ) ( ) Inpatient mortality 38 (3.6%) 16 (1.5%) Any complication 468 (43.7%) 466 (43.6%) Myocardial infarction 12 (1.1%) 11 (1.0%) Hypotension/Shock 25 (2.3%) 13 (1.2%) Pneumonia/Respiratory failure 97 (9.1%) 71 (6.6%) Gastroparesis/Nausea & vomiting 96 (9.0%) 108 (10.1%) Total parenteral nutrition 133 (12.4%) 113 (10.6%) Acute renal failure 58 (5.4%) 36 (3.4%) Urinary tract infection 80 (7.5%) 82 (7.7%) Surgical site infection 117 (10.9%) 128 (12.0%) Blood transfusions 267 (25.0%) 242 (22.6%) DVT/PE 22 (2.1%) 32 (3.0%) who did not. By matching patients with similar probabilities of receiving EA, analogous to randomization, propensity score matching attempts to adjust for this selection bias as well as more evenly distribute any residual confounders not accounted for in the analysis. As with any study utilizing administrative data, there are limitations to this work. The data used in this study were at the population level; thus, individual patient records could not be examined, which prevented us from being able to know the precise details surrounding hospital stays. For example, neither the duration of EA use nor the adequacy of EA with regards to pain control could be examined. Because the NIS was used, this study relied on ICD-9CM codes to identify EA. Thus, it is not known whether patients, who had unsuccessful attempts at EA or had non-functional EA, were included in the EA group. Similarly, where patients were managed post-operatively (ICU versus surgical ward) as well as the types of medications used in EA (i.e. narcotics versus local anaesthetics versus both) are also not included in the NIS. It would also have been beneficial to know how intravenous fluids were managed both intra- and post-operatively as EA has been associated with vasodilatory hypotension, which could result in overzealous fluid administration. As the NIS is limited to inpatient data, it only includes outcomes that occurred during patients initial post-operative stay, and could not account for any complications diagnosed in the outpatient setting or upon readmission. Additionally, some important factors, such as baseline functional status, are not included in the NIS, so there is also the possibility of residual confounding, which could result in incomplete risk adjustment. The involvement of a specialized pain team, which often participates in the post-operative care of patients with EA, is not tracked in the NIS. Therefore, there is a possibility that EA is simply a surrogate for the involvement of a specialized pain team in post-operative management. While propensity score matching is a valid method for analysing non-randomized observational data, it is by no mean perfect. A large prospective randomized trial of EA versus no EA in pancreatectomy patients would be the best method to determine if EA improves outcomes. In conclusion, within a large, administrative hospital database, EA is associated with improved peri-operative outcomes and reduced hospital charges after a pancreatectomy. The finding that EA improves peri-operative outcomes after a pancreatectomy must be interpreted with caution. This analysis, at the national level, may not absolutely apply to any one individual hospital. Additional studies are required to fully understand if this relationship is causal as well as to determine how to best use EA in the setting of a pancreatectomy. Conflicts of interest None declared. References 1. Gendall, KA, Kennedy, RR, Watson, AJ, Frizelle, FA. (2007) The effect of epidural analgesia on postoperative outcome after colorectal surgery. Colorectal Dis 9: ; discussion Jayr C, Beaussier M, Gustafsson U, Leteurnier Y, Nathan N, Plaud B et al. (1998) Continuous epidural infusion of ropivacaine for postoperative analgesia after major abdominal surgery: comparative study with i.v. PCA morphine. Br J Anaesth 81: Mann C, Pouzeratte Y, Boccara G, Peccoux C, Vergne C, Brunat G et al. (2000) Comparison of intravenous or epidural patient-controlled analgesia in the elderly after major abdominal surgery. Anesthesiology 92: Nishimori, M, Low, JH, Zheng, H, Ballantyne, JC. (2012) Epidural pain relief versus systemic opioid-based pain relief for abdominal aortic surgery. Cochrane Database Syst Rev CD

8 558 HPB 5. Popping DM, Elia N, Van Aken HK, Marret E, Schug SA, Kranke P et al. (2014) Impact of epidural analgesia on mortality and morbidity after surgery: systematic review and meta-analysis of randomized controlled trials. Ann Surg 259: Rodgers, A, Walker, N, Schug, S, McKee, A, van Kehlet, H, Zundert, A et al. (2000) Reduction of postoperative mortality and morbidity with epidural or spinal anaesthesia: results from overview of randomised trials. BMJ 321: Coolsen MM, van Dam RM, van der Wilt AA, Slim K, Lassen K, Dejong CH. (2013) Systematic review and meta-analysis of enhanced recovery after pancreatic surgery with particular emphasis on pancreaticoduodenectomies. World J Surg 37: Lassen K, Coolsen MM, Slim K, Carli F, de Aguilar-Nascimento JE, Schafer M et al. (2008) Guidelines for perioperative care for pancreaticoduodenectomy: Enhanced Recovery After Surgery (ERAS(R)) Society recommendations. World J Surg 37: Sanford DE, Olsen MA, Bommarito KM, Shah M, Fields RC, Hawkins WG et al. (2014) Association of discharge home with home health care and 30-day readmission after pancreatectomy. J Am Coll Surg 219: Halabi WJ, Kang CY, Nguyen VQ, Carmichael JC, Mills S, Stamos MJ et al. (2014) Epidural analgesia in laparoscopic colorectal surgery: a nationwide analysis of use and outcomes. JAMA Surg 149: Elixhauser A, Steiner C, Harris DR, Coffey RM. (1998) Comorbidity measures for use with administrative data. Med Care 36: Austin PC. (2007) Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: a systematic review and suggestions for improvement. J Thorac Cardiovasc Surg 134: Austin PC. (2011) Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 10: Austin PC. (2009) Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 28: Pratt WB, Steinbrook RA, Maithel SK, Vanounou T, Callery MP, Vollmer CM Jr. (2008) Epidural analgesia for pancreatoduodenectomy: a critical appraisal. J Gastrointest Surg 12: Shah DR, Brown E, Russo JE, Li CS, Martinez SR, Coates JM et al. (2013) Negligible effect of perioperative epidural analgesia among patients undergoing elective gastric and pancreatic resections. J Gastrointest Surg 17: Popping DM, Van Aken HK, Brodner G, Wenk M. (2014) It is not the epidural that is dangerous, but the person who gives it. Br J Anaesth 112: Stenger M, Fabrin A, Schmidt H, Greisen J, Erik Mortensen P, Jakobsen CJ. (2013) High thoracic epidural analgesia as an adjunct to general anesthesia is associated with better outcome in low-to-moderate risk cardiac surgery patients. J Cardiothorac Vasc Anesth 27: Birkmeyer JD, Siewers AE, Finlayson EV, Stukel TA, Lucas FL, Batista I et al. (2002) Hospital volume and surgical mortality in the United States. N Engl J Med 346: Hyder O, Dodson RM, Nathan H, Schneider EB, Weiss MJ, Cameron JL et al. (2013) Influence of patient, physician, and hospital factors on 30- day readmission following pancreatoduodenectomy in the United States. JAMA Surg 148: